Enigma Raises $70M for Human-Robot Interaction as Multiverse Raises $570M to Shrink Models
Two funding rounds closed the same day at opposite edges of the AI stack. Enigma, which runs large-scale experiments exploring how humans interact with robots, emerged from stealth with a $70 million seed led by Index Ventures and Ribbit. Multiverse Computing, based in Spain and focused on shrinking large language models to reduce energy and compute costs, raised a $570 million Series C at a $1.7 billion valuation.
One is betting that the interface is the unsolved problem. The other is betting that the cost curve is. Both are wagers against the assumption that scaling frontier models is where the remaining value sits.
Why interaction data is worth a $70 million seed
A seed round at that size for a company whose product is experiments rather than robots is unusual enough to explain. The bet is that the constraint on deployed robotics is not manipulation or locomotion — both of which have improved substantially — but the behavioral layer: how a machine signals intent, how it yields in shared space, how it recovers legibly when it fails, and how people calibrate trust in it over repeated exposure.
That data does not exist at scale because it cannot be simulated. Human responses to machine behavior have to be observed with actual humans, in actual settings, which makes the collection apparatus itself the asset. A company that owns a large corpus of interaction data owns something no amount of compute reproduces, and it sits upstream of every robotics firm that eventually needs it.
Ribbit’s participation alongside Index is worth noting given its financial services background. That combination suggests a view of the company as an infrastructure and data business rather than a hardware play.
Why compression attracts nine figures
Multiverse’s raise is the larger and more immediately legible of the two. Inference cost has become the dominant operating expense for anyone deploying models at scale, and memory prices are compounding the problem — DRAM tightness driven by high-bandwidth memory allocation has begun surfacing in consumer hardware pricing, which means the substrate underneath every deployment is getting more expensive at the same time demand for it grows.
Compression attacks that directly. A model that delivers acceptable quality at a fraction of the parameter count runs on cheaper hardware, at lower power, with lower latency, and in places a frontier model cannot go at all — on device, inside a regulated network, at the edge of a factory floor.
The strategic argument is stronger than the cost argument. Every organization that cannot send data to an external API needs capable local inference, and capable local inference is a compression problem before it is anything else. That is a large market defined by constraints rather than by budget.
The common assumption
Both rounds assume the frontier is not where the returns are. Enigma assumes robot capability will arrive and the bottleneck will be human acceptance. Multiverse assumes model capability will arrive and the bottleneck will be affordable deployment.
Those are the two positions available to anyone who believes the core capability race is effectively decided among a handful of well-capitalized labs. If it is, the remaining value accrues to whoever solves distribution and cost — which is the standard pattern in every prior computing platform shift, and the reason these rounds cleared at these sizes.